CME-YOLOv5: An Efficient Object Detection Network for Densely Spaced Fish and Small Targets

Fish are indicative species with a relatively balanced ecosystem. Underwater target fish detection is of great significance to fishery resource investigations. Traditional investigation methods cannot meet the increasing requirements environmental protection and investigation, existing technology has few studies on dynamic identification underwater small targets. To reduce disturbances solve problems many fish, dense, mutual occlusion difficult targets, an improved CME-YOLOv5 network proposed detect in dense groups First, coordinate attention (CA) mechanism cross-stage partial networks 3 convolutions (C3) structure fused into C3CA module replace C3 backbone you only look once (YOLOv5) improve extraction feature information accuracy. Second, three layers expanded four, which enhances model’s ability capture different dimensions improves performance. Finally, efficient intersection over union (EIOU) loss function used instead generalized (GIOU) optimize convergence rate location Based actual image data number datasets obtained online, experimental results showed that mean average precision ([email protected]) algorithm reached 94.9%, 4.4 percentage points higher than YOLOv5 algorithm, performances was 24.6% higher. The show our exhibits good performance when applied densely spaced targets can be as alternative or supplemental method for investigation.